Great news - nVidia GPU folding soon

[The Coolest]

Users of nVidia GPUs, supporting CUDA will soon be able to fold using their video cards, and it should be pretty fast according to this article:
http://www.pcper.com/comments.php?nid=5679

Found it here originally:
http://www.xtremesystems.org/forums/showthread.php?t=188726

[Shiggity]

7000 petaflops going unused!

They also hint at the 'upcoming Nvidia GPU', I'm assuming they mean the new 280GTX :)

+1 to F@H

[WarriorII]

All I have to say is:

It's about darn time.

[NoWhereM]

Sounds good, I hope it will take advantage of sli and that it comes out soon.

[hank123]

sweet our good cards will be put to use!

[harlam357]

Good news for all those running nVidia... I posted about CUDA here, heck, about a year ago... nice to see Vijay and group finally got around to putting it to work. :)

Unless I see some insane ppd numbers... I will stick with my CPUs and save my GPUs. My feeling is that it will be tough to beat the points per watt ratio on a dedicated cpu.

[sno.lcn]

This is the best news I've gotten all day! :attn:


Hooray nvidia! :clap:

http://img520.imageshack.us/img520/8625/hooraynvidiaaf5.jpg

*Too much free time at work :rolleyes:

[Shiggity]

We would have had Nvidia F@H clients much sooner, it just took this long to get Nvidia to agree to it / help. Mostly because Nvidia didn't want it to be only ATI, yay competition.

Another post about it - http://folding.typepad.com/ - looks like it gets a really sweet visualization client like on the PS3 :)

[Super Nade]

One more link stating that Folding on NV GPUs can begin....
http://en.expreview.com/2008/05/24/foldinghome-nvidia-gpu-client-unveiled/

[blakehenry00]

If the Folding@Home client is based on CUBA, that means Folding@Home will be accessible for all the GPU above GeForce 8XXX

[hank123]

whens this going to hit?

[dark bishop]

it says on nvidia gpu's that support cuda, doesnt that mean the tesla and quadro series cards?

[Leviathan41]

Wow, their new card is supposed to fold 5 times faster than the PS3! Don't know if it's exactly right, but still impressive :thup:

As far as exact numbers go, this sucker (the GTX280) can fold at 500 mol/day, which is much higher than the Radeon HD 3870 numbers (170 mol/day), five times higher than PS3 numbers (100 mol/day), and astronomically higher than the average computer numbers (4 mol/day).

http://www.techpowerup.com/index.php?61210

[Sorin]

Wow, their new card is supposed to fold 5 times faster than the PS3! Don't know if it's exactly right, but still impressive :thup:

Yeah. GPUs have gotten pretty ridiculous in terms of the power they're capable of. The raw computational power they have is just stupid (I mean that in the good way). For what? Purely for the sake of making gamers happy. Who would have thought we (gamers) could drive technological advancement in such a beneficial way. :D It's really awesome that the rest of the industry is finally coming around to using the ungodly number crunching power GPUs have. I know the calculations GPUs do are pretty specific (unlike a CPU, which is more general), but it's still very useful.

I'm just waiting for the day the parts on a graphics card become upgradable so that you can build a semi-custom graphics card just like you can build a custom computer. Hell, they're mini computers already as it is. You've got the processor, the memory, the motherboard and a heatsink. Not holding my breath on this one though.

[Shiggity]

Except that you can buy an entire Q6600 folding rig for as much as 1 280GTX :D

[Zerix01]

it says on nvidia gpu's that support cuda, doesnt that mean the tesla and quadro series cards?

Cuda support is on all Geforce cards (consumer level and professional) 8xxx series and up.

This has to do with the stream processors in the G8x G9x chips which have no predefined use (like all cards prior) which means they can be programed for general use floating point calculations.

If they bring this to Linux (not holding my breath) I will buy a second 8800GTS to occupy my lonely second PCI-E slot. The code GROMACS and CUDA are very portable so a Linux version is not out of the question but like all things Folding@Home they need time and resources to get it done.

[Zerix01]

Hmm looking over at the Folding Forums.... it looks like they already have a beta GPU2 client with Nvidia support. Look around a bit and you will see they do plan on releasing a Linux and Mac OS X client once the Windows version is stable.

http://foldingforum.org/viewforum.php?f=42

WOOT

[WarriorII]

So... which Nvidia cards can be Folded?

[Th3 F4ll3n P1mp]

I read about this on XS. Looks pretty promising. Btw, if you guys thought a ps3 racked up points like crazy, wait till you do it on a g92 core. Ps3 is g80 btw. :p

[Zerix01]

I read about this on XS. Looks pretty promising. Btw, if you guys thought a ps3 racked up points like crazy, wait till you do it on a g92 core. Ps3 is g80 btw. :p

PS3 uses a G7x series GPU and the PS3 Folding client only uses the Cell processor, the GPU is not used in the PS3.


So... which Nvidia cards can be Folded?

Opps I guess it isn't released yet but will be supported in a soon to be released beta. The following is a quote from the Folding forums right from the main man himself.

by VijayPande on Fri May 23, 2008 5:35 pm

We should be able to support any card/driver combo that supports CUDA (most G80 cards or later).

[Th3 F4ll3n P1mp]

PS3 uses a G7x series GPU and the PS3 Folding client only uses the Cell processor, the GPU is not used in the PS3.
G7x gpu? Someone said G80 lol. And yeah I know that its only the cpu thats used in the ps3 but I said that just to give an idea of the speed of it lol.

Also, the ps3 cell processor, what cpu on the market would it be about equal to?

[Adak]

There is no equal to the Cell processor, in the PC marketplace.

The reason is that it combines several extremely fast RISC chips, with a front end Complex Processor Unit.

When the 8 core (true independent, integrated cores) comes out, they may be able to keep up with the current Cell processors being used in the PS3.

But the beauty of the Cell is, you can simply add more RISC chips to it, and VROOM! goodbye Mr. 8 core cpu - see ya in about 3 years. :)

[Shiggity]

There is no equal to the Cell processor, in the PC marketplace.

The reason is that it combines several extremely fast RISC chips, with a front end Complex Processor Unit.

When the 8 core (true independent, integrated cores) comes out, they may be able to keep up with the current Cell processors being used in the PS3.

But the beauty of the Cell is, you can simply add more RISC chips to it, and VROOM! goodbye Mr. 8 core cpu - see ya in about 3 years. :)

IBM is already working hard on the 32nm process too :D

Is Cell based on 65nm?

[Shiggity]

woops double post, delete ;)

[TommyHolly]

I have a bunch of older computers with Pentium IV with HT and Nvidia 7600 GT video cards... I know this is a long shot but I really hope that the 7000 series is included somehow?
This would breath new life into these power hungry PCs that only give me 200-400 ppd

I wonder how this will work on my new Quad-core that I just installed a 8800 GTS 512 (G92)? I heard the GPU folding programs need at least one core of the PC to work with the GPU... So would you be running SMP on 3 cores and the 4th would in charge of helping the GPU fold?

[Adak]

I believe the days of the power hungry cpu's is over, and won't return, in any viable way. Vijay may prove me wrong, however. :)

Having an SMP WU run on 3 cores is probably as good as a 3 three legged horse running in the Kentucky Derby.

A better balance might be running 1 smp using VM on 2 cores, then run 2 X GPU clients on SLI, or 1 GPU and one single core processor. Something that will give a (hopefully) balanced and full load to all the cores.

The heat inside the box might be a new high for folding, and a problem for the system, however.

[The Coolest]

I have a bunch of older computers with Pentium IV with HT and Nvidia 7600 GT video cards... I know this is a long shot but I really hope that the 7000 series is included somehow?
This would breath new life into these power hungry PCs that only give me 200-400 ppd

I wonder how this will work on my new Quad-core that I just installed a 8800 GTS 512 (G92)? I heard the GPU folding programs need at least one core of the PC to work with the GPU... So would you be running SMP on 3 cores and the 4th would in charge of helping the GPU fold?
CUDA only supports G8 series and up

[TommyHolly]

What is CUDA? I'm sure that is short for something? Why is CUDA important to Folding?

[Shiggity]

They wrote the folding software into CUDA (It's insanely better using CUDA). It's just a GPU language that's somewhat new that helps the GPU do cpu type calculations. All of the new Nvidia GPU's have it.

http://www.engadget.com/2007/05/25/nvidias-cuda-turns-gpus-into-high-powered-cpus/

OMFG check this out guys - http://www.ocforums.com/showpost.php?p=5647662&postcount=1

Props to Mr. Guvernment for finding it ;)

[Shiggity]

New post on the F@H site about GPU's and their PPDs.

"I've talked about this in the Folding Community Forum (FCF), but I thought it's worth reminding people (especially NVIDIA-based folders new to the project). As the GPU2 code matures, we may need to change the points per day (PPD) and/or the nature of the benchmark machine. The PPD for the bechmark machine may go up or may go down depending on what change we decide in the end. The main problem is that both the NVIDIA and ATI code is undergoing optimizations, which means PPD will change in complex ways. For example, if the NVIDIA code doesn't change, but the ATI code gets optimized and our benchmark is based on an ATI card, then NVIDIA PPD will go down, even if the NVIDIA code doesn't change.

Also, how efficient these cards are depends on the protein we run on a given card. To address this issue, we will likely switch the benchmark machine to run a card whose PPD does not change so variably with protein, although that may be hard to find (we'll see).

Anyway, just a heads up -- the PPD we see today is not necessarily what you'll see later. Your PPD may increase or decrease depending on many complex factors (code optimization, benchmark machine, protein simulated, etc), and I expect there may be more points variability with GPUs than with other platforms (isn't life on the bleeding edge fun). However, due to the great performance of the GPU's right now (which will likely only get better), the GPU PPD would be significantly greater than a typical CPU client. :attn:(A 'typical' cpu client is a P4 probably, but still good news imo)

I should also remind donors that we've been talking about rebalancing points in general, to try to get SMP, GPU, PS3, and classic points in line with the scientific productivity done by the clients. As long as more points = more science, donors can optimize for points and that will automatically optimize for maximum scientific output."

Source - http://folding.typepad.com/

[Sorin]

" ... I should also remind donors that we've been talking about rebalancing points in general, to try to get SMP, GPU, PS3, and classic points in line with the scientific productivity done by the clients. As long as more points = more science, donors can optimize for points and that will automatically optimize for maximum scientific output."

If this is done right, thank you. Even though a GPU can do astronomically more work than SMP, science wise, the SMP units are worth phenomenally more points and more PPD. I don't see it happening the way I envision it though. Even though a GPU does 5-6 times more work than a dual or quad does, I don't see the points for GPU units being 5-6 times higher. That would, unfortunately, probably discourage a lot of people. >.<

Even if the ratio of points/unit to science remains lackluster, I'd still like to see them add some kind of science measurement to the output logs. Currently I've only seen science measurements in the actual console window for the Linux SMP clients (and it isn't logged). I'd like to see these measurements for all clients, and also logged to the text file. I want to see what kind of performance my actual components are pumping out! My E6600, when it was doing VMWare Linux SMP, was doing about 6.6GFLOPS on the 1760 point WUs. I think it was at 3.4GHz. I'd like to know what my quad does and what my GPU would be doing.

[TommyHolly]

WOW! That's crazy? Just one computer running (4) 9800GTXs has the computational power of 300 Core2 Duo PCs... That jumps the whole "Moore's Law" thing quite a bit eh?

[hank123]

arrr I want it now!